Fix(MInference): fix the requirments

app.py

@@ -1,10 +1,10 @@
-# import subprocess
-# # Install flash attention, skipping CUDA build if necessary
-# subprocess.run(
-#     "pip install flash-attn --no-build-isolation",
-#     env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
-#     shell=True,
-# )
+import subprocess
+# Install flash attention, skipping CUDA build if necessary
+subprocess.run(
+    "pip install flash-attn --no-build-isolation",
+    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
+    shell=True,
+)
 
 import gradio as gr
 import os

requirements.txt

@@ -3,5 +3,4 @@ accelerate
 transformers
 wheel
 setuptools
-flash_attn
 pycuda==2023.1